Chat GPT thread

[interstitial]

once again chat gpt fails. It is a new tech toy which might eventually have value but right now it is counterproductive without expert editing. It currently leads to misinformation and should be treated as such. Each use on this forum has led to significant factual errors and IMO people should be discouraged from using it to create forum content.

[oren]

once again chat gpt fails. It is a new tech toy which might eventually have value but right now it is counterproductive without expert editing. It currently leads to misinformation and should be treated as such. Each use on this forum has led to significant factual errors and IMO people should be discouraged from using it to create forum content.

I think I've found my personal chat GPT. Nearly every post I read by interstitial I think to myself: "exactly!"

[Sigmetnow]

once again chat gpt fails. It is a new tech toy which might eventually have value but right now it is counterproductive without expert editing. It currently leads to misinformation and should be treated as such. Each use on this forum has led to significant factual errors and IMO people should be discouraged from using it to create forum content.

How did it fail?  Please, I would like to know.

The information was summarized as presented in the article.  I used a summary to avoid lengthy quotations from the article (which I’ve been dinged for by Neven), and to avoid a lot of repetition of similar information surrounding the Brienz landfall, which was several posts above mine.

I thought people would read the shorter summary, rather than my needing to write HEY, EVERYBODY, THIS IS A DIFFERENT EVENT THAN THE BRIENZ ONE.  I even separated the Brienz information in a different section.

 
Seriously, if you see errors in the summary, please point them out.  It seemed to be an accurate description to me.  I’m testing the summarize function as an option to lengthy quotations.

[kassy]

Well you could try actual concise writing. And yes leading with a simple heads up that this is similar but somewhere else would be a logical starting point.

I think a lot of people don´t like this kind of stuff and a lot of the things it has come up with are not that interesting. Yes it summarizes things but here it is just a list of what always happens. You can easily skip that until people wonder how these things happen but if you read the thread and links you could just add the new report and the most interesting details about it. None of that requires AI power.

 

[Renerpho]

How did it fail?  Please, I would like to know.

I wouldn't call it a "fail". Just a really bad summary. ChatGPT repeats the beginning of the article almost ad verbatim, and completely ignores the remaining 95%. You don't need ChatGPT for such a "summary". You could just have copy/pasted the first two sentences of the article and convey the same information, without the decrease in credibility that comes from using a chat bot.

That would have been perfectly fine, actually, because those initial sentences tell the basic facts, as the beginning of any journalistic article should do. Here's what that would have looked like:

"Part of the summit of a mountain in the Austrian state of Tyrol has collapsed, sending more than 100,000 cubic meters of rock crashing into the valley below and triggering mudslides.
Rocks started falling Sunday from Fluchthorn, a nearly 3,400 meter (11,155 foot) mountain in the Silvretta Alps on the border between Switzerland and Austria, in an incident state geologists have said was caused by thawing permafrost."

Or you could have given your own summary of what you thought was important. That would have been fine, too. Again, ChatCPT can't help you with that, because it can't know what you consider worth mentioning.

If I had tried to post this, I may have pointed to a sentence like this, which I find interesting: "There isn’t enough data to say whether rockfalls have increased in recent years, as it is usually only large events that are well-documented, said Phillips. Scientists rely on information from the public, she added, and many rockfalls happen in remote areas."
Something about the role of citizen scientists! That's nice. Even relevant to this forum, in some sense. There's no way ChatGPT would pick up on this though.

[Sigmetnow]

How did it fail?  Please, I would like to know.

I wouldn't call it a "fail". Just a really bad summary. ChatGPT repeats the beginning of the article almost ad verbatim, and completely ignores the remaining 95%.

Actually, that was me.   I only posted the first part of ChatGPT’s summary — again, for brevity.  Figuring anyone who was interested would click on the link and read the whole thing.

You don't need ChatGPT for such a "summary". You could just have copy/pasted the first two sentences of the article and convey the same information, without the decrease in credibility that comes from using a chat bot.

That would have been perfectly fine, actually, because those initial sentences tell the basic facts, as the beginning of any journalistic article should do. Here's what that would have looked like:

"Part of the summit of a mountain in the Austrian state of Tyrol has collapsed, sending more than 100,000 cubic meters of rock crashing into the valley below and triggering mudslides.
Rocks started falling Sunday from Fluchthorn, a nearly 3,400 meter (11,155 foot) mountain in the Silvretta Alps on the border between Switzerland and Austria, in an incident state geologists have said was caused by thawing permafrost."

But that is almost precisely what I posted, which were the first items in the ChatGPT summary:

Summary by ChatGPT:
 
🔘 Part of the summit of a mountain in the Austrian state of Tyrol has collapsed, sending more than 100,000 cubic meters of rock crashing into the valley below and triggering mudslides.

🔘 The incident happened on Sunday, June 11, on Fluchthorn, a nearly 3,400 meter (11,155 foot) mountain in the Silvretta Alps on the border between Switzerland and Austria.

🔘 The collapse was caused by thawing permafrost, a long-term frozen layer of soil and rocks, according to state geologists.

And then I added the bit about Brienz, which wasn’t in the summary but I though was a relevant update.

Or you could have given your own summary of what you thought was important. That would have been fine, too. Again, ChatCPT can't help you with that, because it can't know what you consider worth mentioning.

 
Of course, and I’m not adverse to that, but it opens the way for human subjectivity — inaccuracy, if you will — which, ironically, seems to be exactly the thing folks fear most about ChatGPT.   

To date, I have found that the ChatGPT summaries (via the app I am using) pretty much just extract sentences from the article.  Thus it is similar to what a human might choose, to put the point across.  After reading the article, I use the summary to save time and help reduce my tendency toward long posts — I don’t just mindlessly copy it verbatim.  And if there is other information that is interesting or important, I add that.

Thus, not much different than a “pure human” post.  But I’ve been labeling the summaries as ChatGPT in case someone finds errors I didn’t catch.  No one has mentioned any, so far.   So… much ado about nothing?

[HapHazard]

posting in a chatgpt thread

[Renerpho]

@Moderator: If there's a thread about ChatGPT, or where this is more appropriate, please move it there. My apologies for discussing it here.

I wouldn't call it a "fail". Just a really bad summary. ChatGPT repeats the beginning of the article almost ad verbatim, and completely ignores the remaining 95%.

Actually, that was me.   I only posted the first part of ChatGPT’s summary — again, for brevity.  Figuring anyone who was interested would click on the link and read the whole thing.

In that case, that criticism wasn't entirely justified. It's still a bad summary because it's not a summary (just a rephrasing), but it's not as bad as I had thought -- because you, the human, did some editing to it. I have to ask though what you needed the summary for? Was it quicker than copy/pasting those sentences yourself? What did the chat bot do that you couldn't have done better?

But that is almost precisely what I posted, which were the first items in the ChatGPT summary: [...]

Yes, that's precisely my criticism (and why I posted it, so you could compare). ChatGPT didn't summarize, it repeated. And the result was less credible than the original, because it came from an unreliable source that's known to make unpredictable errors.

Of course, and I’m not adverse to that, but it opens the way for human subjectivity — inaccuracy, if you will — which, ironically, seems to be exactly the thing folks fear most about ChatGPT.

If there is one thing that I do not fear about ChatGPT, it is human subjectivity. It does hallucinate sometimes, which is a problem, and it is definitely biassed in subtle ways. Both are problems if people use it to access/process information.

Thus, not much different than a “pure human” post. But I’ve been labeling the summaries as ChatGPT in case someone finds errors I didn’t catch.  No one has mentioned any, so far.   So… much ado about nothing?

There wasn't much opportunity for error, because ChatGPT pretty much just took a few sentences from the article, and it did so without committing errors (it sometimes makes strange errors in situations like that, but not in this example). So yes, you could say "much ado about nothing".

I will always prefer a human summary over an automated one. If I want an unbiased version, I read the scientific publication that's behind the press release that's behind the news. I can decide to do that, but often, when I go to this forum and read stuff, I am more interested in what people think about certain issues than the issues themselves. Chat bots don't think anything.

[Sigmetnow]

@Moderator: If there's a thread about ChatGPT, or where this is more appropriate, please move it there. My apologies for discussing it here.

Same here. I can’t seem to find a ChatGPT thread.  But this should be my last post on the subject.

I’m not a cult fan of ChatGPT — I’ve posted several negative articles about it.  And I also look askance when people use it to generate original text created from whatever sources it chooses to use.

But asking it to summarize a specific article (at least, using the app I use to do so) limits its scope to that article — and I’m not asking the AI to reduce a highly technical or complex report.  In fact, the biggest complaint about a ChatGPT summary I’ve gotten so far is that it reads too much like the original article!
https://forum.arctic-sea-ice.net/index.php/topic,2352.msg370803.html#msg370803

As Renerpho noted, ChatGPT mostly pulls direct quotes.  No outside interpretation, other than choosing which quotes to include —  which is the same thing everyone else here does when they post excerpts.

I do read the article first, and decide what, and how much, of the piece to post, or to comment on. ChatGPT merely helps rein in my wordiness. So any feeling that such posts are not “human written” are misplaced.

People are always free to click the link and read the entire article (unless it is paywalled…).  And they are, of course, free to take issue with anything posted here or written in the article, just as with any other post.  If you do notice a discrepancy between the AI notes and the article, I hope you will mention it.  Thanks!

[Bruce Steele]

Sigmetnow, “ ChatGPT merely helps rein in my wordiness. ” I kinda like your posts as they reflect your personality which is a counterpoint to my rather dark soul. There is no soul for a machine although they may leave glimmers of whoever programmed them. The problem with chat IMO is I will never get any insight into another humans soul and words and writing can get cold and lifeless without a human who generally maintains one personality, which gives me some sense of whether I will bother to continue reading what they post. So

[Sigmetnow]

Sigmetnow, “ ChatGPT merely helps rein in my wordiness. ” I kinda like your posts as they reflect your personality which is a counterpoint to my rather dark soul. There is no soul for a machine although they may leave glimmers of whoever programmed them. The problem with chat IMO is I will never get any insight into another humans soul and words and writing can get cold and lifeless without a human who generally maintains one personality, which gives me some sense of whether I will bother to continue reading what they post. So

Thanks, Bruce.  Fear not, I will still expound verbosely as the spirit wills. The ChatGPT summarize app simply offers up quick quotes from the article, saving me the time needed to copy each article off-line and slog through it in order to delete the extraneous bits from my post.

[kassy]

I will split out some of the Chat GPT posts into a separate thread tomorrow.

[trm1958]

Good kassy
Here is another post for the thread
I use ChatGPT to provide summaries of issues for my Toastmasters speeches (but I proofread the results) and also to try to bamboozle it.

[Renerpho]

Here's one more about ChatGPT. I'm sure we can continue the discussion in the new thread. (Thanks Kassy.)

"OpenAI sued for defamation after ChatGPT fabricates legal accusations against radio host"
(from The Verge, 9 June 2023)

Mark Walters, a radio host from Georgia, is suing OpenAI for defamation. A journalist had used the bot to summarize a document. In its summary, ChatGPT wrongly claimed that Walters had stolen in excess of $5,000,000 from a non-profit organisation. This is not true at all.

This is the first legal case about the question of whether a chat bot can commit defamation.

https://www.theverge.com/2023/6/9/23755057/openai-chatgpt-false-information-defamation-lawsuit

ChatGPT didn't actually have access to the article it was asked to summarize, but it did not tell the journalist. Rather, it made up a fictitious summary on the fly, probably from contextual clues given to it. That summary appears to have been convincing enough to pass. An impressive feat in its own right, but quite concerning as well.

[Sigmetnow]

Here's one more about ChatGPT. I'm sure we can continue the discussion in the new thread. (Thanks Kassy.)

"OpenAI sued for defamation after ChatGPT fabricates legal accusations against radio host"
(from The Verge, 9 June 2023) …[/size]

Yes, I posted about this in the Robots and AI thread:
https://forum.arctic-sea-ice.net/index.php/topic,1392.msg370739.html#msg370739

[HapHazard]

posting in a chatgpt thread

[edit] Thanks kassy, that was getting freakin' annoying!

[kassy]

Name changed.

[kassy]

Offer any toy and people will play with it. People loved FaceBook never mind the triangulation.

Bitcoin wasted so much energy but still people had to try that because it promised money.

And now you have some AI but why have it do what you can actually yourself simply within 5 minutes?

This technology gobbles up yet more power which is better used elsewhere.

If it discusses something complicated how do you work out is is right? Something similar goes for all kind of YouTube channels which get thrown around. How do we know they are right and not some waste of energy.

And it is all a finger tap away. So easy to use but should you?

[Sigmetnow]

When choosing excerpts from an article to post, how does one choose what to keep and what to delete? It’s entirely subjective. Some people might choose one thing, some might choose another.  Do you fault them for their choice?

No one has commented that the ChatGPT items I posted were inaccurate, or extraneous.  (Except folks who didn’t really read the quotes and understand they were not repetitious from previous posts.) So the method used to choose them is quite immaterial — except to those people who reject new technology and fear what they don’t understand. 

The app I’m using works directly in my browser, and takes only seconds — not even your five minutes — to generate a representative selection of quotes from an article from which to choose. [After I have read the article and decided it was worth posting, of course.]

<< You can do that by hand, yourself, why would you want a faster, easier way?  If it’s faster and easier, it must be worse. QED! >>

Don’t like quotes selected from those offered by ChatGPT?  Then don’t read them.  Click on the link and spend your time reading the entire article, instead.  The summary might even give you an idea of whether that effort might be worthwhile.

—-
For those interested, the app I’m using is “Ask & Summarize”, from the Apple App Store.  It’s new, and was written by the same programmer who wrote the Stats Tesla app (which not only provides statistics on the Tesla cars it has permission to see, but offers all sorts of nifty notification and activation features, including the ability to control your car from your Apple Watch!  The Stats Tesla app is quite high tech, and he’s continually adding new features, and doesn’t otherwise track your information the way many apps do.  Good stuff.)

“Ask & Summarize” is subscription-only, but you get a free trial. Not mentioned in the app description is that the developer is working on adding summaries for videos — which, judging by kassy’s repeated pleas for video details, would definitely be a big help in deciding if videos posted to the Forum might be worth your time….

[sidd]

Is it too much to ask that those using LLMs in any threads clearly label provenance of the material ?

[kassy]

When choosing excerpts from an article to post, how does one choose what to keep and what to delete? It’s entirely subjective. Some people might choose one thing, some might choose another.  Do you fault them for their choice?

In a way it is but you also post it in a certain thread so as a rule of thumb you can just go with article name + shortest version of what they discovered and a bit of context.

Sometimes i nag about certain choices but that also never helps.

Most are fine but at least people do it themselves.

I really fail to see the need for such an app. If you just read an article it is trivial to copy paste a couple of sentences.

That is the point about toys. Nobody needs them but they are fun. But these things also cost a lot of energy. Yes it can do your homework but it still wastes energy.

The really quick way to check Youtube videos is to read the autogenerated script.
Summarizing that is one thing but then you still do not know how reliable the source is which is often iffy with YT.

So that is all wasting energy.

[Sigmetnow]

When choosing excerpts from an article to post, how does one choose what to keep and what to delete? It’s entirely subjective. Some people might choose one thing, some might choose another.  Do you fault them for their choice?

In a way it is but you also post it in a certain thread so as a rule of thumb you can just go with article name + shortest version of what they discovered and a bit of context.

Sometimes i nag about certain choices but that also never helps.

Most are fine but at least people do it themselves.

I really fail to see the need for such an app. If you just read an article it is trivial to copy paste a couple of sentences.

If the end result is the same quotes, what difference does it make?  Only personal prejudice would make them somehow less acceptable.
 
But if ChatGPT’s quotes highlight items buried further down in the article, that’s a quick way to include more information than copying just the first few sentences, requiring no extra effort to create a more helpful post, with the end result being more informative for people who don’t read the whole article. 

Easy.  Better.

That is the point about toys. Nobody needs them but they are fun. But these things also cost a lot of energy. Yes it can do your homework but it still wastes energy.

The really quick way to check Youtube videos is to read the autogenerated script.
Summarizing that is one thing but then you still do not know how reliable the source is which is often iffy with YT.

So that is all wasting energy.

So we should all use phones instead of tablets to read the Forum, because phones have smaller batteries, and God forbid you use a real computer with a monitor and a metal box that plugs into the wall and sucks so much power it requires an internal fan, plus a clunky keyboard and a mouse on a cable.  For shame!
 
Versus a Chat GPT query that takes about as much energy as a Google Search. Should everyone stop doing that, as well?  We should all remain stupid because using the internet requires energy?

You can read and summarize an hour-long video’s autogenerated script in under a minute?  Wow!

ONE MORE TIME: the app is pulling quotes from the selected article.  The source is right there, just as valuable, and just as accurate — or not — as the original.  YouTube is YouTube, so everything is questionable.  Selecting quotes or summarizing a video is simply an aid for making the choice to spend more time with it, or pass it by.

[Bruce Steele]

Sigmetnow, Every choice we made or make ,every last watt we use adds up. Collectively we are killing the planets life systems. The ocean is in deep trouble already, nobody sees it. Cry for our host planet like you would mourn the loss of your only child.
 Mark the decline in birds, in the insects and marine invertebrates.
 God is coming

[NeilT]

Is it too much to ask that those using LLMs in any threads clearly label provenance of the material ?

Depends on what you are using the LLM for.

[Sigmetnow]

Is it too much to ask that those using LLMs in any threads clearly label provenance of the material ?

I think you can easily distinguish the ChatGPT Summarize sections in my posts.
 
Like this quote:
 
🔘 The search is continuing Tuesday after the Titan lost contact with its mother ship, the Canadian research vessel Polar Prince, during a dive on Sunday morning

But mentioning “ChatGPT” seems to trigger some people — even though it simply pulls quotes from an article just like everyone does here, and no one has said it has presented any erroneous information — so I may not always use that term going forward.

[Sigmetnow]

Sigmetnow, Every choice we made or make ,every last watt we use adds up. Collectively we are killing the planets life systems. The ocean is in deep trouble already, nobody sees it. Cry for our host planet like you would mourn the loss of your only child.
 Mark the decline in birds, in the insects and marine invertebrates.
 God is coming

Spare me.  The choice of using a phone or tablet rather than a desktop computer saves orders of magnitude more watts than doing a Google search or asking ChatGPT to summarize an article.  The micro-second of energy required by a search or a summary is much less than the minutes of battery power required to keep one’s tablet screen lit while scrolling through an article to find, copy and paste quotes manually. 

We could all use less energy if we signed off the Forum permanently, but I have to believe the information provided by posters, including you, helps many others to act more beneficially.

[Bruce Steele]

Sigmetnow, Do you know if there are mainframes burning up lots of kW whether you use them or not?
I just thought Chat would be something like bit coin in burning up huge amounts of energy. I have no idea. If enough people use Chat and use it a lot how will bit coin and Chat compare in use of energy?
Maybe Chat will learn to evade questions regarding it’s energy use so as not to offend people worried about such things.

[sidd]

I asked in an earlier post  that those using LLMs (large language models)  in any threads clearly label provenance of the material. NeilT pointed out that would depend on what one was using the LLM for.

So let me refine my request: Would posters using LLMs to summarize an article please indicate the text generated by LLMs in their posts ?

Sigmetnow stated that one could easily distinguish the ChatGPT Summarize sections in his(hers? its?) posts. Actually i could not because i mostly read this forum in a stripped down, textonly, terminal based browser that i wrote for text to speech for some, let us say visually impaired, colleagues. It wasnt until i looked at the thread in an all mod cons browser that i saw the little icon beginning the LLM sections.

Have a care.

sidd

[Sigmetnow]

I asked in an earlier post  that those using LLMs (large language models)  in any threads clearly label provenance of the material. NeilT pointed out that would depend on what one was using the LLM for.

So let me refine my request: Would posters using LLMs to summarize an article please indicate the text generated by LLMs in their posts ?

Sigmetnow stated that one could easily distinguish the ChatGPT Summarize sections in his(hers? its?) posts. Actually i could not because i mostly read this forum in a stripped down, textonly, terminal based browser that i wrote for text to speech for some, let us say visually impaired, colleagues. It wasnt until i looked at the thread in an all mod cons browser that i saw the little icon beginning the LLM sections.

Have a care.

sidd

I did not realize that.  Can you see any emoji characters at all?

If I edit the ChatGPT Summarize text to suit my fancy, does that still count as an LLM product?

Let’s keep in mind that many internet articles are written or edited by machine, now, so it’s not like there’s a strict differentiation anymore.

[Sigmetnow]

Sigmetnow, Do you know if there are mainframes burning up lots of kW whether you use them or not?
I just thought Chat would be something like bit coin in burning up huge amounts of energy. I have no idea. If enough people use Chat and use it a lot how will bit coin and Chat compare in use of energy?
Maybe Chat will learn to evade questions regarding it’s energy use so as not to offend people worried about such things.

I don’t know the exact energy comparison between an internet query and table screen time.  But mining a Bitcoin requires hours of compute time, versus milli- or micro-seconds for a query, so those are nowhere near equivalent.

Millions (billions?) of internet queries happen every second, so it’s not like server farms are standing by awaiting one ChatGPT request.

[gerontocrat]

I reckon the next stage of chatgpt and its competitors is to know all about you when you formulate a query (and to make Alexa and its equivalents) part of the package. I expect you, like me, have many times clicked yes to the accept cookies button, and even if you haven't the machine knows much of what you did and what you believe.

But knowing you, and your interests and beliefs, makes it logical to provide you with an answer that supports your interests and beliefs. After all, it makes no financial sense to the owners of chatgpt for it to answer you that you are wrong, talking crap. It does make financial sense for chatgpt to give you an answer that satisfies you, reinforces your beliefs, makes you happy.

With something like Alexa built in, no more need to type on a keyboard or mobile. Speak and you will be heard and responded to. No chance then to think about and edit your query. Garbage query, garbage response?

ps: I guess chatgpt answers according to the parameters set by you. I was reading a paper about the effect on increasing glacial melt of the destruction of the Thwaites glacier tongue (which the authors said was minimal). But buried in the paper was a single short sentence that said a similar event on the PIG would have an impact 10 times as great. If I had asked chatgpt for a summary of the paper I doubt the PIG would have got a mention.

pps: Oftimes I have found offtopic rich mines of info by letting the browser wander off into pastures new.

ppps: NeilT has used AI to make program code. I guess he made a program spec as the basis for his request and spent some time making sure the program spec was ok. That's several orders of magnitude different from some offhand words thrown at Alexa.

I guess that if nothing else chatgpt et al will make a significant contribution to the growth of the human idiocracy.

This post was carbon neutral. As of yesterday I've got solar pv on me roof

[Sigmetnow]

If I had asked chatgpt for a summary of the paper I doubt the PIG would have got a mention.

Yes, I’ve had a couple instances where particular bits that I thought were important were not included in the summary items, so I pasted an excerpt from the article in a separate section in the post.  Similarly, not all Summarize items seem important enough to include, so I delete them.  A “summary” by definition does not include all information, so reading the article before posting is important.

Even manually pasting lovingly-selected excerpts will omit things. Human choices can help, or they can make things ridiculously lengthy trying to include everything.  No method is perfect.

But the best use of ChatGPT is as an aid, not a replacement.

[Renerpho]

I suppose we could ask Sigmetnow to stop using ChatGPT. Or make posting AI-generated content a bannable offense. Of course that would be silly!

We may not like it, but using ChatGPT is convenient. I don't personally see the appeal of using it for article summaries specifically, but I trust that those who use it for that purpose, who even defend its use against all opposition, have their reasons. I don't understand it yet, but if there was no practical reason to use it, all that remained was blindly following some trend...

There are some things that I can do with ChatGPT that I couldn't do without it. For example, ChatGPT has all but replaced Google Translate for me, because it's just much better. Do I trust its translations? Well, yeah, a bit more than I trust Google... It's still shit sometimes, but that's what I expect from a machine translation. If I rely on it and get it wrong, I am not surprised, and that limits the use cases. I definitely don't want to become the journalist who makes wrong accusations based on some ChatGPT output.

I think the arguments against its allegedly high energy consumption don't hold up, and are straw men. There are technologies for which this is true, some quite despicable (see BitCoin), but not for LLMs. At least not yet.

For the moment, we are dealing with two issues that are hard to separate.
There's the classic. ChatGPT is a new technology, and like all new technologies, people initially don't trust it. I guess some of the opposition against it on this forum is due to such general fear, which may sound irrational, but shouldn't be laughed off.
But there's an even deeper problem that's new. People's words cannot be trusted to be their own.

When we discuss whether users should clearly mark AI-generated content, one issue is that there are some strong incentives against complying with that. Some people will see your post, read "ChatGPT", and stop reading. Or even worse, get angry and tell you about it. I guess that's generally not the desired outcome, and just not telling anyone that your post wasn't written by you entirely solves that problem.

I'd still argue that we should try to be open about things that we know matter to other people. If we want to collaborate on this forum, it is necessary that we can trust others to play by the rules. Not deliberately deceiving others is part of that.

So. Please, think about why using ChatGPT makes sense when you do it; if some of the content you post is not yours then mark it as such (that's obvious for quotes, maybe less obvious for AI-generated content); and if you find that people don't appreciate it, consider not using those tools anymore, rather than getting angry at other forum members.

Just my opinion. Yours may differ.

[kassy]

Sigmetnow, Do you know if there are mainframes burning up lots of kW whether you use them or not?
I just thought Chat would be something like bit coin in burning up huge amounts of energy. I have no idea. If enough people use Chat and use it a lot how will bit coin and Chat compare in use of energy?
Maybe Chat will learn to evade questions regarding it’s energy use so as not to offend people worried about such things.

I don’t know the exact energy comparison between an internet query and table screen time.  But mining a Bitcoin requires hours of compute time, versus milli- or micro-seconds for a query, so those are nowhere near equivalent.

Millions (billions?) of internet queries happen every second, so it’s not like server farms are standing by awaiting one ChatGPT request.

This is the problem in a nut shell. CharGTP uses more energy then general internet searches but hey so many people are using it.

Not everyone of us though, this is something we play with in the rich first world. Stream things at resolutions devices can´t handle. Have Chat GTP do your homework. It is all extra work and energy and while the internet seems free it is not. Just as buying more crap because everyone is doing it is not free.

We have one planet, one budget. Some use way more energy then others. Maybe think about that?

And aside from that if you prefer to post autogenerated content are you actually in the discussion?

[Sigmetnow]

if you prefer to post autogenerated content are you actually in the discussion?

Posting an article title, a link, and “autogenerated content” to describe that article is no different than what everyone on the Forum does when they add an article to begin, or add to, a discussion.  If personal comments are necessary to add context to the article, we add them!

[Renerpho]

if you prefer to post autogenerated content are you actually in the discussion?

Posting an article title, a link, and “autogenerated content” to describe that article is no different than what everyone on the Forum does when they add an article to begin, or add to, a discussion.  If personal comments are necessary to add context to the article, we add them!

You're right, it's the same thing. If personal comments are necessary then please write them yourself. If they aren't necessary then neither is autogenerated content.

You effectively keep arguing that autogenerated content can replace personal comments, but that's just not true.

[Sigmetnow]

You effectively keep arguing that autogenerated content can replace personal comments, but that's just not true.

I say that to mean my brief summary of an article may not be any more meaningful than a bunch of machine-generated quotes.  It does not refer to my occasional insightful, pithy, or humorous comments.

Other folks may naturally excel at presenting an article in a brief but meaningful way that is superior to simply presenting a few quotes.  That’s awesome, and they should do that!

[Renerpho]

Other folks may naturally excel at presenting an article in a brief but meaningful way that is superior to simply presenting a few quotes.  That’s awesome, and they should do that!

May I suggest that, next time one of your posts includes AI-generated content, you briefly explain *why* you're using it? For example:

"I struggle summarizing articles myself, so here's a summary I generated with help of ChatGPT. I checked that it accurately represents what's in the article: [...]"

There is no shame in using tools. I am quite sure that the response would be considerably less negative. You're creating this summary as a service for the reader, not because, I don't know, you're looking for an excuse to use ChatGPT? And despite the difficulty, you still put effort into it by checking it's accurate. You're not doing it because you're lazy.

If someone still responds telling you that you ought to be able to do it yourself, see if you find a respectful way to tell them to fuck off.

[Sigmetnow]

Other folks may naturally excel at presenting an article in a brief but meaningful way that is superior to simply presenting a few quotes.  That’s awesome, and they should do that!

May I suggest that, next time one of your posts includes AI-generated content, you briefly explain *why* you're using it? For example:

"I struggle summarizing articles myself, so here's a summary I generated with help of ChatGPT. I checked that it accurately represents what's in the article: [...]"

There is no shame in using tools. I am quite sure that the response would be considerably less negative.

My experience suggests such an explanation may be a little too detailed for some folks here to tolerate.  However, I plan to note, “Here’s a few quotes from [or “notes about”] the article,” or some such — if I am using ChatGPT bullet points.  Copying multiple paragraphs from the article itself is common practice here and so when I do, the assumed origin is the correct one.

I still feel that, whether I choose brief statements directly from the article, or I choose them from a ChatGPT selection of them, is immaterial, as long as I verify they accurately represent the article. Because in either case, they are the quotes I chose to post.

[Renerpho]

I still feel that, whether I choose brief statements directly from the article, or I choose them from a ChatGPT selection of them, is immaterial, as long as I verify they accurately represent the article. Because in either case, they are the quotes I chose to post.

I concur.

See if you find an alternative for the "bullet points" that works for those who read this as "text only".

[sidd]

Re: edit the ChatGPT Summarize text to suit my fancy, does that still count as an LLM product?

I suggest the same guidelines as in scientific papers: any more than 15% of contribution should be credited. So if 15% or more of the post is from a LLM, then the LLM ought to be credited as in:

This post was written by [poster] and [LLM]

sidd

[NeilT]

Sigmetnow, Do you know if there are mainframes burning up lots of kW whether you use them or not?
I just thought Chat would be something like bit coin in burning up huge amounts of energy. I have no idea. If enough people use Chat and use it a lot how will bit coin and Chat compare in use of energy?
Maybe Chat will learn to evade questions regarding it’s energy use so as not to offend people worried about such things.

Oddly enough Mainframes do not generally support the internet or even high performance processing.  Mainframes are highly specialised beasts which are highly optimised for managing streams of data and doing merge and transforms on streams of data.

What they are not optimised for is huge amounts of MIPS or TFLOPs for unstructured execution of code or handling unstructured data.  I had this experience back in 2000 when Ericsson bought $40m worth of Mainframes on which they expected to add our portal.  Only to find that the two PIII desktops sitting in our office for testing were more suited to the task than one $10m mainframe.

For AI The base system tends to be x64 based processor clusters (Linux), but layered on top of that are graphics and Neural Network (NN) processors.

There is not a lot of AI compatible hardware going around burning up TW/h doing nothing.  In fact most systems spin down to a very low power state when not being used much.

training an AI requires truly HUGE amounts of power.  However once it is trained, retaining the neural net and maintaining the net plus the storage behind it is not in the same order of magnitued.

In fact extracting data form the neural net could be significantly less power intensive than ploughing through databases of stored data and trying to sort out the hits and matches.  Especially with a very complicated query.

[NeilT]

My position on ChatGPT for summarising an article is similar to using Google for your research.  Firstly Google only finds the matches with the highest number of links to it.  Also you may need to use the exclude "-" terms to even get a result.  Then you also might need to use a time bracket to get the results you need.  For instance if you want to see vaccine research results and analysis which are not polluted by the politics of the Covid era you must cap your searches at Dec 2019.

We all use Google but how many actually subscribe to educational sources and use the library of educational articles?  As opposed to searching in Google where you will find both these articles and also real world advances from industry which never came near an educational institution.

Even then if you use an AI to summarise your article you are going to read the summary and compare it with the article you read.

This then can give you a significant reduction in time to produce your post.  We don't all have unlimited time.

So even when the summaries have been presented by ChatGPT they have been reviewed by a human and compared to the original text.

This is very different from using a chatbot service which leverages ChatGPT on which you give it a topic, your credentials on the site and then tell it to go get articles, get ChatGPT to summaries them and post the results.

There needs to be a clear differentiation between AI "Enhanced" or "Enabled" content and AI generated content.

Most of the resistance here is around AI generated.  Whereas I take it Sig is using AI to either Enable, by reducing time taken, or Enhance by making sure she has not missed stuff she wants to highlight; or a combination of both.

[Sigmetnow]

I still feel that, whether I choose brief statements directly from the article, or I choose them from a ChatGPT selection of them, is immaterial, as long as I verify they accurately represent the article. Because in either case, they are the quotes I chose to post.

I concur.

See if you find an alternative for the "bullet points" that works for those who read this as "text only".

That’s the purpose of prefacing them with, “Here’s a few quotes from [or “notes about”] the article,” or some such — if I am using ChatGPT bullet points.

Perhaps my prior use of the word “summary” has been misleading.
I have not, nor do I have any desire to, post any paragraphs of ‘original’ text that AI has generated — unless I state am doing so for the express purpose of illustrating ChatGPT output.

Re: edit the ChatGPT Summarize text to suit my fancy, does that still count as an LLM product?

I suggest the same guidelines as in scientific papers: any more than 15% of contribution should be credited. So if 15% or more of the post is from a LLM, then the LLM ought to be credited as in:

This post was written by [poster] and [LLM]

sidd

If the bullet points differ a little from the exact text of the article (sometimes a few words is used to summarize a couple sentences), I will label the bullet point section as “notes about,”  “description,” or “points,” rather than “quotes”.  Otherwise, words not in Quote boxes or Titles are mine.

Most of the resistance here is around AI generated.  Whereas I take it Sig is using AI to either Enable, by reducing time taken, or Enhance by making sure she has not missed stuff she wants to highlight; or a combination of both.

Exactly.

—
The “wasted energy” argument is specious, given that posts which merely contain a link are allowed, forcing multiple readers to ask the internet to open the article if they want to read more, whereas the addition of quotes or comments right there in the post may be sufficiently informative to satisfy those who have only a casual (or no) interest in the subject. 
 
And videos of course should have been banned ages ago — think of all the bandwidth they consume!

[Renerpho]

There needs to be a clear differentiation between AI "Enhanced" or "Enabled" content and AI generated content.

Most of the resistance here is around AI generated.

That's an important observation.
It's roughly what I meant when I said it's fine using it as a tool, but not to replace/generate "personal" commentary.

[Freegrass]

I'm not sure where to post this. I just came across this website that looks like a good tool for some of you here. Anyone heard about Scispace before? Here's their intro video. It's free.

https://typeset.io/

Why wait for an author to explain a paper? Say hello to SciSpace, the easiest and fastest way to find, understand, and learn any research paper.

* Search from 270 millon+ papers to discover interesting papers
* Get simple explanations and full context to complex text, equations, and tables
* Ask questions and follow-up queries to get instant answers
* Find connected papers, authors, topics, and more.

Experience a new way of exploring and reading research papers for free:

[Sigmetnow]

ChatGPT forms partnership with German media conglomerate.

Alex @alex_avoigt
 
This is by far the best selection of inaccurate clickbait, smear and FUD media I've ever seen.
If you want to be properly informed about technology, avoid them or believe the opposite of what they report.
 
Good counterindicator.
12/14/23, https://x.com/alex_avoigt/status/1735266218003587237

 
OpenAI @OpenAI
We have formed a new global partnership with @axelspringer and its news products.
 
Real-time information from @politico, @BusinessInsider, European properties @BILD and @welt, and other publications will soon be available to ChatGPT users.
 
ChatGPT’s answers to user queries will include attribution and links to full articles for transparency and further information. openai.com/blog/axel-spri…

12/13/23, https://x.com/openai/status/1734940445824937993
 
< Oh good god. You do know that you just listed 2 of the biggest fake news machines Germany has? Especially BILD is infamous for just making stuff up. Guess I will have to drop ChatGPT. Luckily there are enough alternatives out there.

[SeanAU]

The internet and ....

Living Dead

The nominally coined ‘dead internet theory’ revolves around the idea that bots and AI are turning the net into an artificial goop of endlessly reverberating algorithms, which in essence create a simulacra-simulation matrix of surrogate activity, a sort of human facsimile echo-chamber where actual humans are increasingly left out in the cold looking in, as disenfranchised spectators. One foray into Twitter discourse leaves one with the feeling it may not be far from the truth.

ChatGTP and a variety of other new seemingly near-sentient AI bots are already in widespread use, and are set to overtake the journalism and article writing fields. Everything from companies to content creators are already using these AI for a variety of purposes. A judge in Colombia has used the AI to make a ruling on a case, and the bot is now being regularly used to pen spammy internet articles, social media posts, college entry exams, craft market strategies, and a host of other things.

But the eerier usage results from pairing an AI like ChatGTP with other AI-generated products like avatars and voices, to create a full-on life-like digital facsimile: 

https://darkfutura.substack.com/p/dead-internet

[Freegrass]

Suggestion for a new name of this thread in the tittle of this message.

What I want is an AI that can go through all my messages I've ever written on Facebook and all other social networks, and that can use all those replies and messages to write a book with all my ideas and suggestions for a better world...

How long until this is possible? I have a lot of text on my computer too, that I would love to include...

How long before I can turn all the stupid things I've ever written into something comprehensive and interesting that people would be interested in in reading? 1 or 2 years?

And if so, how many books are gonna get printed when this option becomes available? 🤔

[Freegrass]

Excellent podcast from Bill Gates With Sam Altman.
Look in the video section for some visual snippets, but they don't cover the entire podcast.